millenniumwaraigisfandomcom-20200223-history
Talk:Holy Warrior's Challenge/@comment-24.159.39.167-20161117221750/@comment-28059753-20161119093732
@anon *"And if you start with 6 unlucky events and afterwards have 4 that perfectly match the 50:50 ratio, you will still end up believing in a 80% chance." Huh? This is perfectly rational. If you believe there is a fixed drop rate, and you have a uniform prior on the different possible drop rates then 80% is the most likely in the posterior. (Or I would say 20% but that is assumed to be semantics.) Though you seem to be talking about dice or coins here, we're ultimately talking about an event with some unknown characteristic(s) if we're trying to apply such arguments to game drops. *"That's how things work and how statistics trick people. History doesn't matter for it. Just because an event rarely happened in the past does not mean higher chances in the future. Thats a common misunderstanding of the law of large numbers." (Not sure if same user since IP is different, but I thought so.) This is true for independent events (though not in general). If we're talking about the fixed drop rate model, then we are assuming the drop events are independent, but under Ssvb's model, drop events are not independent. Doesn't matter though, as this seems to miss the whole point anyway. If we're talking about an unknown rate, then additional data can and should influence our belief distribution (assuming we are not 100% certain -- like I said unknown rate). That doesn't mean we have to believe the drop rate changed or that there was correlation between the events. (If the data can't influence the belief, then there is no point in testing! We are just stuck with whatever belief we already have whether it is right or wrong.) It does seem you may have a strong prior belief in the 50% rate. Even if I agreed, I wouldn't suggest using such a prior in statistical calculations, but rather only keeping it in mind for the strength of evidence needed against such a value (if indeed there is any such evidence). *"Yes, I just flipped a coin 6 times. Head, head, tail, head, tail, tail. I mean, what are the chances of that? Something like 1,6% I guess." Ssvb has the right idea in response to this. Sure, all coin flip sequences of a specific length are equally (un)likely. This is completely missing the point. The point is that by identifying parameters of interest (e.g. number of Maribel drops), it is perfectly valid to perform tests of this parameter and then do statistical analysis of the results. The reason they are special as compared to whatever random sequence you propose is that they were identified beforehand as the object of the test. This is the elementary idea of statistical hypothesis testing that Ssvb has been harping about. It is perfectly rational to see that one's own results (or the results of others) have some deviation from the proposed model (e.g. exactly 50% drop rate) and to interpret this as evidence against the proposed model. In fact, this is correct. It **is** evidence against the proposed model. However, it's generally very weak evidence due to the small number of runs, and many of these people overvalue their evidence. And such evidence can reasonably be considered biased by other observers -- hence the importance of tests that control sampling bias (if one cares about such things at all). Additionally, many of these people who post such claims seem to have little or no understanding of statistics. And, yes, such people are frequently annoying. So what? I have already tried to argue against Ssvb's idea at considerable length. I think it has lots of big problems. It doesn't mean everything Ssvb is saying is completely wrong.